Spec before modelling: the document that saves you three weeks

I have shipped three commercial modular environment kits. Two of them went smoothly. One of them took roughly three weeks longer than it should have, and I can point at the exact reason: I started modelling before I could answer the question “how many times does this piece appear in a scene?”

That question sounds trivial. It is not. It determines the polycount you can afford, whether the piece gets a unique texture or shares a trim sheet, whether it snaps to a grid, and — the expensive one — whether it needs to work at three different scales or one. Answer it wrong on a 116-mesh kit and you find out at the point where fixing it means redoing UVs on forty assets.

So now I write a spec first. It is one page. It takes an afternoon. It is the single highest-leverage hour in the whole project.

Why “just start blocking out” fails specifically for kits

Blocking out and iterating is genuinely good advice for a scene. A scene has one camera, one composition, one lighting setup, and you find the answers by looking. A kit has none of those things. It has an unknown buyer building an unknown scene, and your job is to make every reasonable arrangement work.

The failure mode is subtle. Everything looks fine while you are building the first fifteen pieces, because you are unconsciously arranging them the way you imagined them. The trouble arrives when you try the twentieth arrangement — the one your buyer will try — and discover that your wall pieces only line up if the corner goes on the left.

This is the part where I should say what actually happened to me.

The spec, in four decisions

Not a template to fill in mechanically — four questions that each rule out a lot of downstream work.

1. What is the module length, and what owns the grid?

Pick one base length and derive everything from it. I use 200 cm for building kits, because it divides cleanly into a 100 cm sub-module and sits comfortably against Unreal’s default 100 cm grid, so a buyer can snap pieces without changing their editor settings.

Then decide, explicitly, which piece owns the origin. My rule now: corners own the grid, walls fill between them. Corners are the pieces with two constraints, so they get to be right and the walls absorb the tolerance.

That last row is worth stating in your store listing, not just your spec. “Snaps on the default grid” removes a real objection for a buyer who has been burned by a kit built on 137 cm modules.

2. How many times does each piece appear?

Sort every planned asset into three buckets. This one table decides your whole texture budget.

The mistake — and it is the most common mistake in kits generally — is treating every prop as a hero prop. If a piece appears twenty times, a unique 2K map for it is twenty times the memory for one instance of visual interest. Put that budget into the one piece the camera actually stops on.

3. What goes on the trim sheet, and what does not?

Once you know the frequency buckets, the trim sheet almost writes itself: every structural piece has to live on it. The decision that remains is how many trims, and the honest answer is fewer than you want.

My working rule is one trim sheet per material family, not per asset type. The cabin kit has two: painted timber, and stone-and-mortar. Metal fittings share the timber sheet because there are only four of them and they are small.

A trim sheet plan is a drawing, not a list. Draw the sheet at the start, with the strips labelled and the UV space allocated, before you author a single texel. If the strips do not fit, you have too many, and finding that out on paper costs ten minutes instead of two days.

4. What is the texel density, and is it the same for everything?

One number, written down, applied to the whole set. I use 512 px/m for kits like this, which is the common game standard and holds up until the camera is closer than about a metre.

Consistency matters more than the value. A kit where every piece is 512 reads as professional; a kit where some pieces are 1024 and some are 256 reads as careless even if the average is fine, because the mismatch is visible exactly where two pieces meet.

If you are not sure what number your surface sizes imply, the texel density reference page has a calculator that will tell you in about four seconds.

What the spec is not

It is not a design document and it should not be pretty. Mine is a single page of plain text and one hand-drawn trim sheet layout. Nobody else reads it. Its whole job is to force the four decisions above into an explicit form so that when I am eleven assets deep and tempted to fudge something, there is a written answer to argue with.

It is also not immutable. Change it when you learn something — but change the document, not just the model. The document being stale is how you end up with two conventions in one kit.

The one-page template

Here is the actual shape of mine, stripped to headings. Copy it, delete what does not apply, and fill it in before you open anything.

KIT: <name>
TARGET: <engine version, renderer features you rely on>
THEME: <one sentence — what a buyer would search for>

GRID
  base module          <cm>
  sub-modules          <cm>
  vertical module      <cm>
  grid owner           <corners | walls>
  pivot convention     <...>
  snaps on default grid?  <yes/no>

TEXEL DENSITY
  target               <px/m>
  exceptions           <which pieces, and why>

TRIM SHEETS
  sheet 1              <material family>  <resolution>
  sheet 2              <material family>  <resolution>
  drawn layout         <link to the sketch>

ASSET LIST
  structural (20+ uses)    <count>   trim only
  character  (3-8 uses)    <count>   trim + small atlas
  hero       (1-2 uses)    <count>   unique set

BUDGETS
  triangle ceiling, structural piece   <n>
  LOD count                            <n>
  collision                            <simple / per-piece / hull>
  total texture memory target           <MB>

DELIVERABLE
  formats              <UE, FBX, glTF...>
  demo level?          <yes/no>
  documentation        <what the buyer gets>

OPEN QUESTIONS
  <the things you genuinely do not know yet — keep this list honest>

That OPEN QUESTIONS section is the most useful part. Everything you are tempted to hand-wave goes there, visibly, and you resolve it before it becomes a refit.

Next

Part two goes deeper on the single decision that caused my three-week refit: grid discipline. Which pieces must sit on the grid, which must not, and how to tell the difference before you find out the expensive way.